Feature – Designed, built and accepted on its own
M1

AI chat moderation

Not started

Sync algorithmic first-pass filter (blocklist + anti-bypass), async AI judge in the background, severity tiers, auto-mute, held messages, retroactive report escalation, appeal path

Phase

Phase 4

Size

L

Waits on

2

Releases

0

Story – A unit of delivery work, sized to land in one sitting

Stories

Not designed yet.

Stories and acceptance criteria are written one Feature at a time. Nothing is missing – this one has not had its pass.

Decisions of record

  • Chat Needs design

    AI-assisted moderation – architecture confirmed, model and severity detail still needs design. A synchronous, cheap first-pass algorithmic filter (blocklist plus anti-bypass normalisation) blocks the obvious stuff inline, in the send path. Every message that passes it still fires an async call to an AI judge in the background – the model is never in the critical path of a send. Severe judgements auto-mute after the fact. Borderline judgements are held silently; if that message is later reported, the prior judgement escalates the report and can auto-issue the mute retroactively.

History

  • 3 Sept Not started – Seeded from the delivery plan, 2026-09-03.